Bookmark · 1 May 2023, 12:09 The Dual LLM pattern for building AI assistants that can resist prompt injection mlgrokme